Loading...
Please wait, while we are loading the content...
Similar Documents
A Flexible Resource Co-Allocation Model based on Advance Reservations with Rescheduling Support
| Content Provider | CiteSeerX |
|---|---|
| Author | Netto, Marco A. S. Buyya, Rajkumar |
| Abstract | Several parallel and distributed applications require simultaneous access to resources located in multiple administrative domains. Current research on resource co-allocation relies on either rigid advance reservations or non-booking-in-advance mechanisms. The first approach leads to high fragmentation inside the resource provider’s scheduling queue, whereas the second approach offers no starting time guarantees of user applications. In this work, we propose a new model for resource co-allocation based on flexible advance reservations and processor remapping. The model allows the metascheduler to reschedule the co-allocation requests by modifying the starting time of each subtask and remapping the number of processors used by them in each resource provider. We evaluate our model and algorithms in a scenario where users are not able to provide accurate runtime estimations of their applications—using job response time and system utilization as metrics. The results show that rescheduling co-allocation requests brings benefits for both local and multi-site applications especially when the runtime estimation quality is low and there is a reduced number of small jobs in the system. 1 |
| File Format | |
| Access Restriction | Open |
| Subject Keyword | Rescheduling Support Advance Reservation Flexible Resource Co-allocation Model Resource Provider Co-allocation Request Runtime Estimation Quality Resource Co-allocation Relies Second Approach Flexible Advance Reservation Resource Co-allocation Multiple Administrative Domain Accurate Runtime Estimation Starting Time Multi-site Application Rigid Advance Reservation Current Research Starting Time Guarantee Several Parallel Small Job First Approach Reduced Number Simultaneous Access Processor Remapping Non-booking-in-advance Mechanism Job Response Time High Fragmentation User Application System Utilization New Model Distributed Application |
| Content Type | Text |
